Reproduce “Agent Laboratory: Using LLM Agents as Research Assistants” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Samuel Schmidgall, Yusheng Su, Ze Wang, Ximeng Sun, Jialian Wu, Xiaodong Yu, Jiang Liu, Michael Moor and 2 more

Agent Laboratory turns a researcher's idea into experiments and a report using agents for literature review, planning, coding, and interpretation, with optional human checkpoints. Co-pilot papers received higher external overall scores than autonomous papers: 4.38/10 versus 3.8/10. Different backends and study settings prevent attributing the difference to feedback alone; the system assists rather than replaces researchers.

Key insights

  • Human-AI collaboration, particularly in a "co-pilot" model, is effective for improving the quality of AI-generated research outputs.
  • Automated evaluation of research quality often overestimates results compared to human assessment, underscoring the need for human validation.
  • The choice of LLM backend significantly impacts research outcomes, with `o1-preview` and `o1-mini` performing better for specific tasks than `gpt-4o`.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.